CN105808382A - Identification and recovery method of abnormal data of transformer substation on the basis of waveform coefficient - Google Patents

Abstract

The invention discloses an identification and recovery method of the abnormal data of a transformer substation on the basis of a waveform coefficient. The identification and recovery method comprises the following steps: firstly, obtaining the practical sampling value of an electrical quantity of a current moment, and meanwhile, predicting a prediction sampling value of the current moment in real time according to the practical sampling value of the electrical quantity in a past period of time; then, comparing the practical sampling value of the current moment and the prediction sampling value of the current moment, and judging whether the practical sampling value of the current moment is possibly the abnormal data or not through a difference value of the practical sampling value of the current moment and the prediction sampling value of the current moment; subsequently, according to a consistent degree between a fitting signal of the practical sampling value of the current moment and the practical sampling value of one subsequent period of time and a waveform theory expression which can represent the electrical quantity signal of a power system, determining whether the practical sampling value of the current moment is abnormal or not, and obtaining the starting point and the end point of the abnormal data; and finally, according to a judged abnormal data point, carrying out next-step work, selecting simple latch-up protection, or selecting practical sampling point data before and after the abnormal data to recover the abnormal data.

Description

Transformer station's disorder data recognition and restoration methods based on form factor

Technical field

The present invention relates to digital transformer substation field, be specially the transformer station's disorder data recognition based on form factor and restoration methods.

Background technology

Along with the construction of China's intelligent grid, the digitized degree of transformer station is more and more higher, and in transformer station, the main medium of information exchange is become optical fiber from cable, and the signal wherein transmitted also is become digital quantity from analog quantity.Electronic mutual inductor and merging unit are widely applied in digital transformer substation; primary voltage and current signal are gathered by e-book transformer and are converted to digital signal, pass to follow-up measurement and protection device processes after merging unit collects, synchronizes.In the process, due to the instability of the interference of external electromagnetic environment and electronic equipment itself, the electric parameters signal of transmission may distortion, show as the sudden change of one or more data point, these data points are referred to as exceptional data point.Abnormal data is not the correct reflection of an electric signal, but the quality factor position in its Frame is normal, measures and protection device is regarded it as normal data and processed, can affect greatly to result, can cause the malfunction of protection time serious.So, secondary device is before processing the sampled value signal received, it is necessary to whether data point judges extremely, and when necessary abnormal data is repaired, to ensure the reliability of secondary data.

About identification and the restoration methods of transformer station's abnormal data, domestic practitioner had done some researchs.The paper " power system electric parameters abnormal sample value real-time identification method " being published in " Automation of Electric Systems " magazine proposes " 3 continuous effective diagnostic methods of sampled value ", analyze that power system is properly functioning and the feature of electric current, voltage waveform under fault, point out that waveform can be led continuously and the same zonal cooling of derivative at other arbitrfary point places except some discontinuous points, and whether sampled value is abnormal to utilize this characteristic to judge.The method sensitivity when abnormal data and normal data deviation are less is not enough, and the abnormal data that None-identified continuous print fluctuating margin is little.Another section of paper " method of the anti-abnormal data of digital transformer substation " being published in " Automation of Electric Systems " magazine proposes a kind of sampled value anti-abnormal data method of detection based on amplitude com parison.The method cannot effectively identify the abnormal data that absolute value is less, and during for electric network fault, will have certain time delay ability open and protection, it is possible to can part quickly protection be impacted.

Accepting with in invention disclosed patent, with the order of magnitude of adjacent two sampling number evidences, " intelligent substation flying spot data processing method " judges that whether it is abnormal by comparison object sampling number evidence, and recover abnormal data by the method for curve matching.The method is less at abnormal data absolute value or will lose efficacy when there are continuous abnormal data." electric power system alternating current magnitude of current sampled data validation checking method " is proposed by continuous three samples value to calculate the quick amplitude of the fundametal compoment magnitude of current, by the quick amplitude of different sampled point calculating place compare mutually and this amplitude and fixing threshold value relatively judge that whether data abnormal.The quick amplitude absolute value that the method effective premise calculates when being system jam less than the quick amplitude of threshold value and abnormal data more than threshold value.It is true that the size of abnormal data cannot be determined, sampled point absolute value when its absolute value is likely to the system failure belongs to the same order of magnitude it could even be possible to less than the latter, for these abnormal datas, the method is None-identified then.

In sum; the absolute value that existing disorder data recognition method commonly uses sampled value compares, continuous sampling point single order or second differnce value relatively judge that whether data abnormal, also exist threshold value be difficult to choose, abnormal data point value less time None-identified, continuous multiple spot disorder data recognition difficulty and affect the problems such as quick operating time of protection.

Summary of the invention

For problems of the prior art, the present invention provides a kind of transformer station's disorder data recognition based on form factor and restoration methods, it is possible to abnormal data being made and effectively identifies and recover, desired data window is shorter, identifies quickly accurately.

The present invention is achieved through the following technical solutions:

Based on transformer station's disorder data recognition and the restoration methods of form factor, comprise the following steps:

Step one, obtains the actual sample value y of electric parameters signal current time_k, the prediction samples value y ' according to the electric parameters actual sample value real-time estimate current time in the past period simultaneously_k；

Step 2, the prediction samples value y ' to current time_kActual sample value y with current time_kCompare and judge whether this actual sample value is likely to be abnormal data；

If judging, the actual sample value of this current time is normal, then perform step one；

If judging, the actual sample value of this current time is likely to be abnormal data, then using the sampled point of current time as potential abnormal data starting point, and perform step 3；

Step 3, the fitted signal according to the actual sample value of current time with the actual sample value in a period of time afterwards, judge that whether current time actual sample value is abnormal with the similarity degree of the electric parameters signal in ideally power system；

If judging, this current time actual sample value is normal, and previous moment actual sample value is normal, then remove potential abnormal data starting point, and perform step one；

If judging, this current time actual sample value is abnormal data, then updating subsequent time sampling number evidence is current time sampling number evidence, and repeat step 3, until the current time actual sample value after updating is normal, then using potential abnormal data starting point as abnormal data starting point, using the current time actual samples point after renewal as abnormal data end point, and perform step 4；

Step 4, according to the abnormal data starting point judged and end point, selects to carry out latch-up protection, or selects according to the sampling number before and after abnormal data according to abnormal data is recovered.

Preferably, in step one, the prediction samples value y ' of current time_kCan be calculated by equation below:

y′_k=-y_k-4+2y_k-3-2y_k-2+2y_k-1+2cosωT_s(y_k-3-2y_k-2+y_k-1)；

In formula, y_k-4～y_k-1For the actual sample value of before current time four sampled points, ω is power frequency component angular frequency, T_sFor the sampling period.

Prediction samples value y ' preferably, in step 2, to current time_kActual sample value y with current time_kThe Rule of judgment compared and judge is shown below:

$\frac{\left|y_k-y_k^{\′}\right|}{I_m}>{\mathrm{\ϵ}}_1;$

In formula, I_mFor power system nominal current magnitude, ε₁Threshold value is compared for prediction；

If the formula of Rule of judgment is unsatisfactory for, then judges that the sampled value of this current time is normal, return step one；If the formula of Rule of judgment meets, then using the sampled point of current time as potential abnormal data starting point, and perform step 3.

Preferably, in step 3, the method for discrimination of electric parameters signal similar degree is specific as follows:

Step 3.1, chooses the actual sample value y of current time_kN-1 sampled value after it calculates form factor R as data window,

$R=\frac{\underset{k=1}{\overset{N-4}{\Σ}}|-y_k+2y_{k+1}-2y_{k+2}+2y_{k+3}-y_{k+4}+2{\mathrm{cos\ωT}}_s(y_{k+1}-2y_{k+2}+y_{k+3})|}{\underset{k=1}{\overset{N-4}{\Σ}}\left|y_k\right|};$

In formula, N calculates the sampled point number used in form factor, and span is N >=5, and N is integer；

Step 3.2, if R is < ε₂Set up, ε₂For form factor threshold value, then judge y_kIt not abnormal data, perform step one；If R is < ε₂It is false, then judges y_kFor abnormal data starting point, and update sampling number according to repeating method of discrimination, until formula R < ε₂Set up, it is determined that sampled point now is abnormal data end point, has namely recorded one section of continuous print exceptional data point, and has performed step 4.

Further, the integer between N desirable 5 to 41.

Further, ε₂Value determined by following steps:

First, choosing the N number of sampled point of continuous print in a segment standard power frequency component and, as the data segment location chosen, wherein first sampled value will replace with abnormal data, the degrees of offset of abnormal data is by ε₁Determine, the form factor expression formula defined calculate the form factor R of this section of sample values；

Then, same segment standard power frequency component changes the data segment location chosen and again chooses N number of sampled point, make the position of first sampled point travel through the whole signal period in repeating the process chosen, calculate different R values, take wherein minimum R and be assigned to ε₂。

Preferably, in step 4, when selecting according to the sampling number before and after abnormal data according to when abnormal data is recovered, restoration methods is as follows:

It is y that note recovers post-sampling value "_k, computing formula is:

y″_k=-y_k+4+2y_k+1-2y_k+2+2y_k+3+2cosωT_s(y_k+1-2y_k+2+y_k+3)；

In formula, y_k+1～y_k+4The actual sample value of four sampled points after the sampling instant corresponding for recovering post-sampling value, ω is power frequency component angular frequency, T_sFor the sampling period；To be restored complete after, remove all records, return step one.

Compared with prior art, the present invention has following useful technique effect:

1, judge that whether data are abnormal according to the fitting effect of destination sample point sampled point follow-up with it rather than its amplitude size, sampling number evidence when can effectively distinguish abnormal data with the system failure.

2, the disorder data recognition of continuous multiple spot can be applicable to.

3, can accurately arranging threshold value according to the patient abnormal data departure degree of system, the sensitivity of algorithm is not by the impact of the size of abnormal data own, and the abnormal data that logarithm value is less also can accurately identify.

4, algorithm desired data window is shorter, 5 sampled points of the shortest need, and the data window length corresponding when being 4kHz of sample rate is 1ms.Window length of generally fetching data is 5ms, it is sufficient to meet requirement movement time of quickly protection.

5, can simply latch-up protection after disorder data recognition, it is possible to as required it is recovered.

Accompanying drawing explanation

Fig. 1 is the flow chart of method described in present example.

Fig. 2 is the schematic diagram of the embodiment of the present invention, and wherein (a) represents the typical current waveform of period before and after electric power system fault, wherein contains different types of abnormal data；B () represents that this algorithm is to the identification situation of abnormal data in (a), " 1 " represents the sampled point of correspondence position is abnormal data, and " 0 " represents without exception；C () is the signal waveform after abnormal data is recovered.

Detailed description of the invention

Below in conjunction with specific embodiment, the present invention is described in further detail, and the explanation of the invention is not limited.

The invention discloses a kind of transformer station's disorder data recognition based on form factor and restoration methods, it is as follows that the method realizes process: first, obtain the actual sample value of current time electric parameters, simultaneously the prediction samples value according to the electric parameters actual sample value real-time estimate current time in the past period.Then the prediction samples value of the actual sample value of current time and current time is compared, judged by both deviation sizes whether the actual sample value of current time is likely to be abnormal data.Next, the fitted signal of the actual sample value according to current time and the actual sample value in a period of time afterwards and the consistent degree that can represent power system electric parameters signal waveform theoretical expression determine that whether the actual sample value of current time is abnormal, and obtain starting point and the end point of abnormal data.Next step work is carried out, it is possible to select latch-up protection simply, it is also possible to select according to the actual samples point data before and after abnormal data, abnormal data to be recovered finally according to the exceptional data point judged.Sampling number evidence when the present invention can effectively distinguish abnormal data with electric power system fault；Disorder data recognition suitable in continuous multiple spot；Can accurately arranging threshold value, the abnormal data that logarithm value is less also can accurately identify；Desired data window is short, meets the quick-action requirement of protection.

Concrete, the transformer station's disorder data recognition based on form factor and restoration methods that the present invention proposes comprise the following steps:

Step one, updates the actual sample value y of current time_k.Meanwhile, the actual sample value y according to before current time four sampled points_k-4～y_k-1The actual sample value of current time is predicted, sampled point and sampled value one_to_one corresponding, it was predicted that formula is as follows:

y′_k=-y_k-4+2y_k-3-2y_k-2+2y_k-1+2cosωT_s(y_k-3-2y_k-2+y_k-1)(1)

In formula, ω is power frequency component angular frequency, T_sFor the sampling period, if the data being combined unit output judge, according to working standard, T_sGenerally take 0.25ms.

Step 2, the prediction samples value y ' to current time_kActual sample value y with current time_kCompare and judge whether this actual sample value is likely to be abnormal data, it is judged that condition is:

$\frac{\left|y_k-y_k^{\&prime;}\right|}{I_m}>{\mathrm{\&epsiv;}}_1---\left(2\right)$

In formula, I_mFor power system nominal current magnitude, ε₁Comparing threshold value for prediction, this threshold value can according to the different electric power patient abnormal data adaptive change of value departure degree.If electromagnetic environment residing for electrical secondary system is comparatively severe, the abnormal data of appearance is of a great variety, ε₁Can being set to a smaller value, for instance 0.5, namely the prediction samples value of current time meets more than 0.5 times of up-to-date style (2) condition of power system nominal current magnitude with the deviation of the actual sample value of current time, starts follow-up processing flow.When environment is ideal, ε₁A higher value can be set to, if the actual sample value degrees of offset of current time is less, without further process.

If formula (2) is unsatisfactory for, then judges that the actual sample value of current time is normal, return step one.If formula (2) meets, the actual sample value sequence number k of current time is assigned to potential abnormal data starting point startpoint, now there are two kinds of possibilities, one is the actual sample value exception of current time, two is power system owing to failure and other reasons produces transient process, and point corresponding to the actual sample value of current time is power system by the stable state critical point to transient state transition.Differentiation for both of these case is completed by subsequent step.

With the fitted signal of the actual sample value in a period of time afterwards and the similarity degree of the electric parameters signal in ideally power system, step 3, judges that whether current time actual sample value is abnormal according to the actual sample value of current time.Ideally the electric parameters signal expression in power system is as follows:

In formula, A is the amplitude of power current,For power current initial phase angle, B is the initial value of attenuating dc component, and τ is the damping time constant of attenuating dc component.

Choose the actual sample value y of current time_kN-1 sampled value after it has N number of sampled value altogether as data window to calculate form factor R, and computing formula is as follows:

$R=\frac{\underset{k=1}{\overset{N-4}{\&Sigma;}}|-y_k+2y_{k+1}-2y_{k+2}+2y_{k+3}-y_{k+4}+2{\mathrm{cos\&omega;T}}_s(y_{k+1}-2y_{k+2}+y_{k+3})|}{\underset{k=1}{\overset{N-4}{\&Sigma;}}\left|y_k\right|}---\left(4\right)$

And R is done following judgement:

R < ε₂(5)

If formula (5) is false, then update the actual sample value of current time, repeat the calculating of formula (4) and the judgement of formula (5).When formula (5) is set up, sequence number k-N+1 is assigned to abnormal data end point endpoint.If startpoint and endpoint represents same sampled point, illustrate that normal transient process is being experienced in now power system, it does not have abnormal data occurs；If startpoint and endpoint represents different sampled points, then the sampled point illustrated between these 2 is exceptional data point.

In formula (4), when sample rate is 4kHz, the integer between N desirable 5～41, algorithm time delay now is 1ms～10ms, it is sufficient to meet the quick-action requirement of protection.

ε₂Value by N and ε₁Together decide on.Choosing the N number of sampled point of continuous print in a segment standard power frequency component and, as the data segment location chosen, wherein first sampled value will replace with abnormal data, the degrees of offset of abnormal data is by ε₁Determine, formula (4) calculate the form factor R of this section of sample values.Then, same segment standard power frequency component changes the data segment location chosen and again chooses N number of sampled point, make the position of first sampled point travel through the whole signal period in repeating the process chosen, calculate different R values, take wherein minimum R and be assigned to ε₂。

When the span of N is 5～41, namely sample rate is the algorithm time delay (data window length) during 4kHz is 1ms～10ms, ε₁Span when being 0.1～1, ε₂Value determined by table 1.

Table 1 ε₂Value

Step 4, is determined the position of abnormal data by startpoint and the endpoint in step 3, normal data points thereafter abnormal data is recovered, and it is y that note recovers post-sampling value "_k, computing formula is:

y″_k=-y_k+4+2y_k+1-2y_k+2+2y_k+3+2cosωT_s(y_k+1-2y_k+2+y_k+3)(6)

Namely from time sequencing, the exceptional data point of last appearance is first recovered, then the exceptional data point occurred before recovering successively.To be restored complete after, remove all records, return step one.

In conjunction with Fig. 2, the effectiveness of this method is illustrated.The sample rate used is 4kHz, and algorithm realizes each parameter in process and selects as follows: I_m=10, ε₁=0.5, N=21, table 1 obtain ε₂=0.026.

Fig. 2 (a) represents that power system is broken down the current waveform before and after the moment, and the signal waveform expression formula of sampled point 1～240 correspondence is:

Y=10sin (100 π t) (7)

The signal waveform expression formula of sampled point 241～480 correspondence is:

$y=50sin(100\mathrm{\&pi;}t-\frac{\mathrm{\&pi;}}{2})+50e^{-10t}---\left(8\right)$

The value changing some sampled point becomes exceptional data point, has 8 places, and the position number often locating abnormal data is as shown in table 2 with Exception Type.

Table 2 abnormal data position number and type

Fig. 2 (b) represents that this method is to the identification situation of abnormal data in (a), and " 1 " represents the sampled point of correspondence position is abnormal data, and " 0 " represents without exception.Can be seen that, except No. 101 sampled points are owing to departure degree is less than ε₁Outside unrecognized, all the other various types of exceptional data points are all effectively recognized.Meanwhile, the sampled point of fault moment be not recognized as abnormal data.

Fig. 2 (c) identifies the recovery situation after exceptional data point for this method, it can be seen that result is ideal.

Claims (7)

1. based on transformer station's disorder data recognition of form factor and restoration methods, it is characterised in that comprise the following steps:

Step one, obtains the actual sample value y of electric parameters signal current time_k, the prediction samples value y ' according to the electric parameters actual sample value real-time estimate current time in the past period simultaneously_k；

Step 2, the prediction samples value y ' to current time_kActual sample value y with current time_kCompare and judge whether this actual sample value is likely to be abnormal data；

If judging, the actual sample value of this current time is normal, then perform step one；

If judging, the actual sample value of this current time is likely to be abnormal data, then using the sampled point of current time as potential abnormal data starting point, and perform step 3；

Step 3, the fitted signal according to the actual sample value of current time with the actual sample value in a period of time afterwards, judge that whether current time actual sample value is abnormal with the similarity degree of the electric parameters signal in ideally power system；

If judging, this current time actual sample value is normal, and previous moment actual sample value is normal, then remove potential abnormal data starting point, and perform step one；

If judging, this current time actual sample value is abnormal data, then updating subsequent time sampling number evidence is current time sampling number evidence, and repeat step 3, until the current time actual sample value after updating is normal, then using potential abnormal data starting point as abnormal data starting point, using the current time actual samples point after renewal as abnormal data end point, and perform step 4；

Step 4, according to the abnormal data starting point judged and end point, selects to carry out latch-up protection, or selects according to the sampling number before and after abnormal data according to abnormal data is recovered.

2. the transformer station's disorder data recognition based on form factor according to claim 1 and restoration methods, it is characterised in that in step one, the prediction samples value y ' of current time_kCan be calculated by equation below:

y′_k=-y_k-4+2y_k-3-2y_k-2+2y_k-1+2cosωT_s(y_k-3-2y_k-2+y_k-1)；

In formula, y_k-4～y_k-1For the actual sample value of before current time four sampled points, ω is power frequency component angular frequency, T_sFor the sampling period.

3. the transformer station's disorder data recognition based on form factor according to claim 1 and restoration methods, it is characterised in that the prediction samples value y ' in step 2, to current time_kActual sample value y with current time_kThe Rule of judgment compared and judge is shown below:

$\frac{|y_k-y_k^{\&prime;}|}{I_m}>{\mathrm{\&epsiv;}}_1;$

In formula, I_mFor power system nominal current magnitude, ε₁Threshold value is compared for prediction；

If the formula of Rule of judgment is unsatisfactory for, then judges that the sampled value of this current time is normal, return step one；If the formula of Rule of judgment meets, then using the sampled point of current time as potential abnormal data starting point, and perform step 3.

4. the transformer station's disorder data recognition based on form factor according to claim 1 and restoration methods, it is characterised in that in step 3, the method for discrimination of electric parameters signal similar degree is specific as follows:

Step 3.1, chooses the actual sample value y of current time_kN-1 sampled value after it calculates form factor R as data window,

$R=\frac{\underset{k=1}{\overset{N-4}{\&Sigma;}}|-y_k+2y_{k+1}-2y_{k+2}+2y_{k+3}-y_{k+4}+2{\mathrm{cos\&omega;T}}_s(y_{k+1}-2y_{k+2}+y_{k+3})|}{\underset{k=1}{\overset{N}{\&Sigma;}}\left|y_k\right|};$

In formula, N calculates the sampled point number used in form factor, and span is N >=5, and N is integer；

Step 3.2, if R is < ε₂Set up, ε₂For form factor threshold value, then judge y_kIt not abnormal data, perform step one；If R is < ε₂It is false, then judges y_kFor abnormal data starting point, and update sampling number according to repeating method of discrimination, until formula R < ε₂Set up, it is determined that sampled point now is abnormal data end point, has namely recorded one section of continuous print exceptional data point, and has performed step 4.

5. the transformer station's disorder data recognition based on form factor according to claim 4 and restoration methods, it is characterised in that the integer between N desirable 5 to 41.

6. the transformer station's disorder data recognition based on form factor according to claim 4 and restoration methods, it is characterised in that ε₂Value determined by following steps:

First, choosing the N number of sampled point of continuous print in a segment standard power frequency component and, as the data segment location chosen, wherein first sampled value will replace with abnormal data, the degrees of offset of abnormal data is by ε₁Determine, the form factor expression formula defined calculate the form factor R of this section of sample values；

Then, same segment standard power frequency component changes the data segment location chosen and again chooses N number of sampled point, make the position of first sampled point travel through the whole signal period in repeating the process chosen, calculate different R values, take wherein minimum R and be assigned to ε₂。

7. the transformer station's disorder data recognition based on form factor according to claim 1 and restoration methods, it is characterised in that in step 4, when selecting according to the sampling number before and after abnormal data according to when abnormal data is recovered, restoration methods is as follows:

It is y that note recovers post-sampling value "_k, computing formula is:

y″_k=-y_k+4+2y_k+1-2y_k+2+2y_k+3+2cosωT_s(y_k+1-2y_k+2+y_k+3)；

In formula, y_k+1～y_k+4The actual sample value of four sampled points after the sampling instant corresponding for recovering post-sampling value, ω is power frequency component angular frequency, T_sFor the sampling period；To be restored complete after, remove all records, return step one.